In order to realize the potential of highly parallel shared-memory MIMD architectures for solving very large problems, novel challenges must be faced by the system software designer. The operating system must endeavor to utilize all processors fully, without incurring serial bottlenecks during coordination operations. Critical code sections far too short or infrequent to cause performance penalties on today's machines will be of concern on very large machines because the cost of each serial section rises linearly with the number of processors involved. Further, the control software must provide basic facilities to support a structured and natural style of general-purpose parallel programming. We present the approaches taken for satisfying these requirements in the NYU Ultracomputer design. We also describe our current preliminary parallel operating system, derived from UNIX, which is currently running on an eight-processor prototype Ultracomputer.
|Original language||English (US)|
|Number of pages||10|
|State||Published - 1985|
ASJC Scopus subject areas